Changes in Extreme Air Temperatures in One of South America’s Longest Meteorological Records: Campinas, Brazil (1890-2022)

Abstract

The frequency of extreme weather events has increased in almost all regions of the world. In this context, it is vital to understand how the location, scale, and shape of their frequency distributions are changing over time. This study used the flexibility of neural networks to model changes in the probability of daily extremes of maximum (Tmax) and minimum (Tmin) air temperature data in Campinas, Brazil — one of South America’s longest meteorological records spanning from 1890 to 2022. Based on the Extremal Types Theorem, we employed a conditional density network to model the parameters of the generalized extreme value distribution (GEV-CDN) as functions of time. Our findings indicate that a GEV-based model, where the location and scale parameters vary over time, best described the variations in the Tmin series. However, a GEV-based model with only the location parameter varying over time best described the variations in the Tmax series. From an agrometeorological perspective, these results suggest that the probability of Tmax values leading to crop failures is rapidly increasing. The findings indicate a decrease in the probability of agronomic frost events in Campinas over the past 133 years, but the rate of this decrease has slowed in recent years. This result, combined with the negative value of the GEV’s shape parameter, suggests that it is unlikely that Campinas may become an agronomic-frost-free region. To facilitate visualization of the changes in the probability of Tmax and Tmin values from 1890 to 2022, we have developed an internet application available at https://climatology-iac.shinyapps.io/Shinny/.